Globalisation and national trends in nutrition and health ... · Globalisation has substantially...

WP 16/18

Globalisation and national trends in nutrition

and health - a grouped fixed effects approach

to inter-country heterogeneity

Lisa Oberländer, Anne-Célia Disdier & Fabrice Etilé

August 2016

http://www.york.ac.uk/economics/postgrad/herc/hedg/wps/

Globalisation and national trends in nutritionand health - a grouped fixed effects approach

to inter-country heterogeneity

Lisa Oberländer∗ Anne-Célia Disdier† Fabrice Etilé‡

July 15, 2016

∗INRA-PSE, [email protected]†INRA-PSE, [email protected]‡INRA-PSE, [email protected]

We are grateful for financial support from the INRA Metaprogrammes Did’it and Glofoods.

1

Abstract

This paper estimates the effect of globalisation on nutritional composition ofthe diet and health outcomes using a panel dataset of 70 countries spanning42 years (1970-2011). Our key methodological contribution is the applicationof the grouped fixed effects estimator developed by Bonhomme and Man-resa (2015), which enables us to better control for unobserved time-varyingheterogeneity. Our results indicate that a one standard deviation increase inthe index of social globalisation is associated with an increase in the shareof animal protein of about 12%. In contrast, economic globalisation has noeffect on the composition of the diet. Moreover, we do not find significanteffects on diabetes prevalence or mean Body Mass Index. Our findings in-dicate that social aspects of globalisation, such as food advertising, deservegreater attention in the nutrition transition discourse.

Keywords: nutrition transition, globalisation, overweight, grouped fixed effects

2

1 INTRODUCTION

1 Introduction

Globalisation has substantially altered food systems around the world, yet conse-quences for nutrition and health are not well understood. This paper estimates therelationship between globalisation, food supply, and health outcomes for a largesample comprising 70 high and middle income countries between 1970 and 2011.Not only has food supply significantly increased over the past 50 years but thecomposition of the diet has been undergoing a profound shift.

Diets have become less dominated by carbohydrates1 while intake of animalprotein, animal fat, and free fat2 have been rising. While this nutrition transitionhas advanced the most in high income countries, the transition speed is even fasterin upper middle income countries. Figure 1 shows that the share of animal proteinrose by 16% (from 6.4% to 7.4%) in high income countries and by 35% (from 3.6%to 4.8%) in upper middle income countries at the expense of vegetable proteins.Similarly, figure 2 reveals that vegetable fat is increasingly replaced by free fat andanimal fat. Free fat rose by 18% (from 14.7% to 17.4%) in high income countries,by 60% (from 8.3% to 13.3%) in upper middle income countries, and by 46%(from 6.4% to 9.3%) in lower middle income countries. The share of animal fatalso increased but to a smaller extent than free fat. Finally, although carbohydratesexperience an overall decline, the share of sugar in total food supply has beenslightly increasing in lower middle income countries (see figure 3).The nutrition transition has been associated with augmenting consumption levels ofultra-processed foods3 and studies document that middle and low income countriesexperienced the fastest growth in processed food consumption during the past 15years (Monteiro et al., 2013; Stuckler et al., 2012).

The nutrition transition constitutes an important risk-factor for non-communicablediseases such as cardiovascular diseases and diabetes. High intakes of fat and sugarcontribute to overweight, which in turn is an important risk factor for diabetes andcardiovascular diseases (CVD) (WHO, 2015b). In 2012 17.5 million people diedfrom cardiovascular diseases ranking them as the number one cause of death glob-ally. More than three quarters of CVD deaths take place in low- and middle-incomecountries causing substantial economic costs (WHO, 2015a). As a consequence,the World Health Organisation (WHO) regards obesity and related diseases as a

1 Carbohydrates are sugars, fiber, and starches found in fruits, grains, and vegetables. Productsrich in carbohydrates are cereals, pasta, rice, bread, corn, peas, and lentils.2 We classified oil, butter, and cream as free fats, as these are not part of a food item but individualscan choose the quantity of free fats in their diet.3 Ultra-processed foods are made from processed substances extracted from whole foods andcontain a high share of unhealthy types of dietary fat, sugar, and salt (Monteiro et al., 2013)

3

1 INTRODUCTION

Figure 1: Composition of protein supply by income group, 1970-2011

Figure 2: Composition of fat supply by income group, 1970-2011

growing threat all over the world replacing traditional public health concerns suchas undernutrition and infectious diseases (WHO, 2000).

Globalisation has been held responsible for the nutrition transition (Hawkes, 2006;Popkin, 2006; Bishwajit et al., 2014). The existing evidence for this claim, however,consists mostly of case studies linking observed changes in diets to specific freetrade agreements (Hawkes and Thow, 2008; Thow and Hawkes, 2009; Thowet al., 2011) and trends in foreign direct investments (FDI) in the food industry(Hawkes, 2006). These case studies typically solely focus on economic aspects ofglobalisation and fail to take into account the multifaceted nature of globalisation.However, theoretical work from Olivier et al. (2008) suggests that it is crucial toanalyse economic and social facets of globalisation and its impact on nutritionalbehaviours separately. More precisely, their model predicts that social integrationcauses cultural convergence while economic integration causes cultural divergence.

4

1 INTRODUCTION

Figure 3: Composition of carbohydrates supply by income group, 1970-2011

The intuition behind this result is that economic integration allows countries toproduce their own food at lowest costs while social integration triggers a shift intaste.

Considering model predictions of Olivier et al. (2008), we use the KOF Index ofglobalisation constructed in Dreher (2006) which allows us to separately analysethe effect of the social and economic dimension of globalisation. In order to bettercontrol for unobserved time-varying heterogeneity we use the Grouped FixedEffects (GFE) estimator developed by Bonhomme and Manresa (2015). The GFEestimator endogenously groups countries together that share similar time profilesof food supply. Thereby, we can control for time-varying unobserved heterogeneitythat is common within groups of countries. Furthermore, we test whether socialand economic globalisation have heterogeneous impacts across country groups.

Our results suggest that social globalisation has a positive and significant effecton the share of animal protein in food supply. In contrast, economic globalisationhas no significant effect on the composition of the diet. The magnitude of theeffects is considerable, as a one standard deviation increase in the index of socialglobalisation is associated with an increase in the share of animal protein 12%.We further find a relatively strong convergence trend across groups for sugar andto a lesser extent for animal protein. Moreover, we find that the effect of socialglobalisation on animal protein is stronger for richer compared to poorer countries.Turning to health outcomes, we do not find significant effects of social and eco-nomic globalisation on diabetes prevalence or mean Body Mass Index (BMI).

Our paper contributes to the following strands of literature.First, this paper is most closely related to a recent study from Costa-i-Font andMas (2014) who estimated the effect of globalisation on calorie intake and obesity

5

1 INTRODUCTION

using data from 26 mostly OECD countries from 1989 until 2004. Results oftheir pooled ordinary lest-squares (OLS) regression suggest that globalisation ispositively associated with calorie intake.We make a methodological contribution by applying the grouped fixed effectsestimator, which allows us to better control for unobserved time-varying hetero-geneity than pooled OLS. For example, food advertising expenditure is likely tobe correlated with globalisation (e.g. number of households with television) andhas been found to be positively associated with absolute calorie intake (Folkvordet al., 2016; Pettigrew et al., 2013). Moreover, we contribute to this literature byanalysing the effect of globalisation on the composition of the diet instead of theabsolute calorie intake. While undoubtedly excess calorie intake increases therisk of obesity, the changing composition of the diet has independent effects onmorbidity and mortality.

Second, this paper relates to the literature on the relationship between overweightand globalisation at the country-level. Three studies (Costa-i-Font and Mas, 2014;Miljkovic et al., 2015; Vogli et al., 2014) use country-level data on BMI andoverweight between 1980 and 2008. Goryakin et al. (2015) pooled DemographicHealth Surveys and restricts its sample to women. Using country fixed effectsregressions4 these studies conclude that the effect of social globalisation is positiveand significantly larger in magnitude than the effect of economic globalisation.Results on the effect of economic globalisation are mixed. Two studies (Vogliet al., 2014; Miljkovic et al., 2015) report a positive linear effect. But Goryakinet al. (2015) documents a small negative significant effect for women. Thesecontradicting findings are likely to stem from different samples, time periods, and(non-)inclusion of individual covariates.We add to the literature by better controlling for time-variant unobserved hetero-geneity. Moreover, to the best of our knowledge this is the first study analysing theeffect of globalisation on diabetes prevalence.

The rest of the paper is structured as follows. Section two describes data and sectionthree the estimation strategy. Results and discussion are presented in sections fourand five. Section six concludes.4 Costa-i-Font and Mas (2014) used pooled OLS

6

2 DATA

2 Data

2.1 Food supply data

Given that we observed the strongest trends for animal protein, free fat, and sugar,we restrict our attention to these three outcome variables. These dietary componentsare also particularly associated with negative health outcomes. High intake of sugarincreases the risk of type two diabetes and overweight (Imamura et al., 2015;Te Morenga et al., 2013). Animal fat and free fats are associated with increasedrisk of coronary heart disease mortality (de Souza et al., 2015; Leren, 1968), andanimal protein elevates the risk of type two diabetes (Malik et al., 2016).The per capita food supply, expressed in kilocalories (kcal) per day, is a measureof the average number of calories available for human consumption, including allfood groups. We obtained these data from the food balance sheets of the FAO forthe time period of 1961 until 20115. A food balance sheet indicates total supply byreporting the total quantity produced of each basic food item, adjusted for imports.On the utilisation side, a distinction is made between quantities exported, fed tolivestock, used for seed, losses during storage and transportation, and food supplyavailable for human consumption. However, the amount of food actually consumedmay be lower depending on the degree of losses in the household.In addition to the kilocalories available for human consumption of each food item,the dataset also contains the amount of fat and protein (grams/capita/day) of eachfood item, which we subsequently converted into kcal/capita/day.In a second step, we determined for each food item its dominant type of fat andprotein. More precisely, we divided proteins into vegetable and animal proteinsaccording to their source. For fat, we differentiated between animal and vegetableorigin and distinguished these from free fats that are not bound in a product. Inparticular, we classified vegetable oils, fish oil, butter, and cream as free fats.Finally, we separated sugar from other carbohydrates and subsequently calculatedthe shares of fat, protein, sugar, and their subgroups in total food supply.

2.2 Health outcomes

For health outcomes we focus on diabetes prevalence and BMI. Data was obtainedfrom the Global Burden of Metabolic Risk Factors of Chronic Diseases Collaborat-5 We drop data for the years 2012 and 2013 because they contain a large number of missing values.Access to FAO balance sheet data: http://faostat3.fao.org/download/FB/FBS/E

7

2.3 KOF Index of globalisation 2 DATA

ing Group, which is a worldwide network of clinical and public health researchers6.The dataset covers the time period of 1980 to 2008 and is constructed by collectingdata from health examination surveys and epidemiologic studies. The researchersused a Bayesian hierarchical model to estimate mean BMI and diabetes prevalenceover time, by age group, sex, and country. Final data is age-standardised corre-sponding to the 2000-2025 world population (Finucane et al., 2011). The BMI is asimple index of weight-for-height that is commonly used to classify overweightand obesity in adults. It is defined as the weight in kilograms divided by the squareof the height in metres (kg/m2). Diabetes is defined as having a mean fastingplasma glucose value of 7.0 mmol/L or greater, or use of a glucose-lowering drug.

2.3 KOF Index of globalisation

Globalisation is a global process including "economic integration, transfer ofpolicies across borders, transmission of knowledge, [and] cultural stability" (Al-Rodhan and Stoudmann, 2006). We use the KOF Index of globalisation developedby Dreher (2006)7, as it allows us to distinguish between the social and economicdimension of globalisation. This is important in order to test the model predictionsof Olivier et al. (2008).

The variables social and economic globalisation take values on a scale from 1 to100 and higher values indicate a higher level of globalisation. Economic global-isation consists of two sub-dimensions: Data on actual flows (1) which includestrade, foreign direct investments, portfolio investment, and income payments toforeign nationals. The second sub-dimension consists of data on trade openness(2) measured by an index of hidden import barriers, mean tariff rate, taxes oninternational trade, and capital account trade openness.Social globalisation is also constructed as a composite index with three sub-dimensions. It contains data on personal contacts (1) measured by telephonetraffic, transfers, share of foreign population, and international letters. The secondsub-dimension is data on information flows (2) including Internet users per day,television, and trade in newspapers. The final sub-dimension of social globali-sation is data on cultural proximity (3) consisting of the number of McDonald’srestaurants and Ikea stores as well as trade in books.

Over the past 40 years globalisation has intensified across the world. Figure 4shows that while high income countries score highest on the globalisation index,

6 We thank the research group for sharing their data. Access to data:https://www1.imperial.ac.uk/publichealth/departments/ebs/projects/eresh/majidezzati/healthmetrics/metabolicriskfactors/.7 We thank the KOF team for sharing their data. Access to data: http://globalisation.kof.ethz.ch/.

8

2.4 Descriptive Statistics 2 DATA

all countries experience a sharp upwards trend of globalisation since the 1990s.Economic and social globalisation has been converging in high income countriesand their scores have been almost identical since the mid 1990s. Middle and lowincome countries still have a higher score in economic than in social globalisationand these two dimensions have evolved parallel over time.Interestingly, we observe a parallel increasing trend for both dimensions of globli-sation while the model of Olivier et al. (2008) predicts that economic and socialglobalisation have opposing effects on the diet.Social and economic globalisation exhibit a high correlation of about 0.84 andthe sub-dimensions cultural proximity and information flows are also stronglycorrelated (0.74). In order to rule out that coefficients are unstable because ofmulticollinearity, we checked the variance inflation factor (vif). All of our variablesexhibit a vif substantially smaller than the rule of thumb of 10.

Figure 4: Social and economic globalisation by income group, 1970-2011

2.4 Descriptive Statistics

Our estimation sample includes 2940 observations from 70 countries. We droppedall countries containing missing values in any of the outcome variables or covariates.Consequently, this balanced sample contains 42 observations for every country,from 1970 until 2011 and covers 76% of the worldwide population. 40% of thecountries in our sample are high income countries and 60% are middle incomecountries.

Table 1 reports the descriptive statistics for all variables. On average 25% of dietaryenergy is derived from fat, about 11% from protein, and 63% from carbohydrates.Free fat accounts for 12%, animal protein for about 5%, and sugar for about one

9

2.4 Descriptive Statistics 2 DATA

fifth of total food supply.In our sample, the mean score of economic globalisation is 52 points and higherthan social globalisation (44 points) on a scale from 0 to 100. Mean GDP per capitais about 10,000 USD.

In order to estimate the effect of globalisation on health outcomes, the samplehas to be reduced to the time period 1980 to 2008 resulting in a sample size ofN = 24948. In our sample around 8% of the population suffer from diabetes. MeanBMI is 24.5, which is slightly lower than the cut-off point of >= 25 for overweightas defined by the WHO (WHO, 2015b).

Table 1: Summary Statistics, nutrition sample

Mean SD Min MaxOutcome variables

Protein in food supply 11.11 1.50 7.68 16.89Fat in food supply 25.64 8.26 7.37 45.25Carbohydrates in food supply 63.25 9.27 41.78 82.43Animal protein in food supply 4.95 2.37 1.05 13.68Free fat in food supply 12.48 5.26 0.84 28.85Sugar in food supply 21.52 6.15 3.85 40.82

CovariatesSocial globalisation 44.06 20.65 6.83 92.31

Personal contacts 48.61 19.43 8.81 90.61Information flows 51.65 21.11 4.40 97.83Cultural proximity 31.53 30.58 1.00 95.95

Economic globalisation 52.14 17.17 17.27 97.09GDP per capitaa 10.50 13.22 0.14 69.09

Income groupsHigh income 0.41 0.49 0 1Upper middle income 0.27 0.44 0 1Lower middle income 0.31 0.46 0 1

SampleNumber of countries 70Number of years 42N 2940

a GDP per capita (constant 2005 in 1000 USD).

8 see table 6 in the Appendix for summary statistics of the smaller health outcome sample.

10

3 ESTIMATION STRATEGY

3 Estimation strategy

3.1 Grouped Fixed Effects (GFE) estimator

Our objective is to estimate the relationship between globalisation, the compositionof the diet, and health outcomes. The main challenge for our identification strategyare unobservable country characteristics that affect a country’s level of globalisationas well as its food supply. For example, cultural norms and unobserved trends infood innovation may both affect a country’s openness as well as its dietary habits.A common solution to this problem is the application of country and year fixedeffects that control for time-invariant unobservable country characteristics as wellas time trends common for all countries. This approach implies the relativelystrong assumption that all unobserved country characteristics are constant overtime and all countries follow the same time trend.

A less restrictive approach is the grouped fixed effects (GFE) estimator devel-oped by Bonhomme and Manresa (2015). The GFE estimator relaxes the strictassumption that all countries follow the same time trend and it only requires thatall countries within a group follow the same time pattern. Yet, different groups canhave distinct time patterns.Consequently, the GFE estimator is better suited to control for unobserved time-varying country characteristics that are correlated with globalisation as well as withcomposition of food supply. In contrast to the country fixed effects estimator, theGFE estimator can control for unobservable time-varying country characteristicsthat follow a group-specific time pattern. The main identifying assumption is thatthe number of distinct country-specific time patterns of unobserved heterogeneityis equal to the number of groups. In other words, all countries have to follow oneof the group specific time-varying paths of unobserved heterogeneity.

For our research question, the GFE estimator constitutes an attractive alternativeto a country fixed effects model. It allows for time-varying unobservables ina time period in which many countries have experienced a nutrition transition.Since transitions in the diet of societies are likely to be triggered by developmentsaffecting several countries at the same time, it is plausible to assume that there areclusters of countries sharing a similar transition profile. For example, in the early1990s, countries formerly belonging to the "Eastern Block" became independent,opened their countries and became exposed to "Western diet" at roughly the sametime. Similarly, expansion of supermarkets in the developing world occurredin several waves starting in the 1990s. The first wave hit major cities in richercountries in Latin America, followed by East and South-east Asia, poorer countriesin Latin America and Asia, and finally South Asia (Reardon et al., 2003).

11

3.2 Estimation procedure 3 ESTIMATION STRATEGY

The second important feature of the GFE estimator is that group membership ofthe countries in our sample is not pre-determined (e.g classification according toincome groups) but that group membership is estimated according to a least-squarescriterion. Countries whose time profiles of the outcome variable (food supply) -net of the effect of covariates - are most similar are grouped together. Assumethat the countries in our sample are categorized in a number of groups indexed byg = 1, ...,G. The number of groups g must be small compared to the number ofcountries. Our regression equation takes the following form:

yit = β1globalisationit +β2GDPpcit +β3(GDPpcit)2 +αgit + vit ,

where yit denotes the outcome variables for country i and year t. In particular, wehave three outcome variables, the share of animal protein (1), free fat (2), and sugar(3) in total food supply.Our coefficient of interest is β1 indicating the effect of globalisation on foodsupply while controlling for GDP per capita and GDP per capita squared9. We usetwo main indicators for globalisation, namely social globalisation and economicglobalisation. In addition, we further disentangle the social dimension and use thethree sub-dimensions of social globalisation, personal contacts (1), informationflows (2), and cultural proximity (3). αgit denotes the group-specific time fixedeffect which includes group fixed effects as well as time fixed effects. vit denotesthe error term.

3.2 Estimation procedure

Bonhomme and Manresa (2015) propose two algorithms for sorting the countriesinto groups. Algorithm 1 is a clustering algorithm and consists of two alternatingsteps.

Assignment stepIn the beginning, a starting value of the parameter values (θ 0,α0) is chosen.Countries are sorted into groups by minimizing the sum of residuals over all yearsfor each country i:

gi = argmingε{1,...,G}∑Tt=1 (yit− xitθ −αgt)

2

In the case of 42 years and 2 groups, for each country the residuals are computed 42times while assuming the country is sorted into group 1 and hence using αg1t . Theresiduals across all 42 years are then summed and compared to the sum obtained

9 The GDP per capita variable is obtained from the World Development Indicators databasepublished by the World Bank. Access to data: http://data.worldbank.org/data-catalog/world-development-indicators.

12

3.2 Estimation procedure 3 ESTIMATION STRATEGY

when repeating this exercise using αg2t , that is assuming that this particular countryhad been sorted into group 2 instead of group 1. Finally, the country is sorted intothat group in which it achieved the smallest sum of residuals over these 42 years.This step is called the assignment step and results in an initial grouping gs

i wheres = 0.

Update stepIn the update step the initial grouping g0

i is used to estimate a new set of coeffi-cients (θ s+1,αs+1). Then, s is set to s = s+1 initializing a new assignment step.Algorithm 1 thus alternates between an assignment step and an update step. Thisloop stops when the difference between the old and the new coefficients is closeto zero. The drawback of algorithm 1 is that the solution depends on the initialstarting value. In order to ensure a reliable solution the entire exercise is simulateds times by choosing s different starting values to begin with. This can result in verylong computation times.

Given this drawback of algorithm 1 Bonhomme and Manresa (2015) also proposethe more efficient algorithm 2 that uses the variable neighbourhood search method.First, a starting value of the parameters (θ 0,α0) is chosen and algorithm 1 is usedto obtain an initial grouping of the countries γinit . The key feature of algorithm 2 isthe inclusion of a neighbourhood jump, where n countries are randomly reallocatedto n randomly selected groups to obtain a new grouping γ

′. These random jumps

allow for an efficient exploration of the objective function and also avoid to gettrapped in a local minimum. The newly obtained grouping γ

′is then used to

perform an update step to obtain new parameter values (θ′,α′). Thus, algorithm 2

cycles over an update step, an assignment step, and a neighbourhood jump. Thiscycle is performed neighmax times where neighmax is set by the researcher. Theentire exercise is simulated Nsim times by choosing Nsim different starting values.For each starting value, the cycle described above is repeated itermax times bysetting n back to 1 once neighmax has been reached.

We use algorithm 2 for our main result and following Bonhomme and Manresa(2015) we set neighmax, Nsim, and itermax all equal to 10. Group-specific coefficientsare obtained with algorithm 1 with 500 simulations10. In order to account for thefact that group membership has been estimated the variance covariance matrix iscomputed by using bootstrapping with 25 replications11

10 To this end we converted the Matlab code "Heterogenous_coeffB.m" provided by Bonhommeand Manresa (2015) into a Stata do file, which is shared on request.11 Bootstrapped standard errors are obtained by setting neighmax = 10,Nsim = 5, and itermax = 5.

13

3.3 Choice of the number of groups 4 RESULTS

3.3 Choice of the number of groups

Bonhomme and Manresa (2015) acknowledge that "optimal choice of G in practiseis a notoriously difficult problem" (p.23 of the supplementary Appendix). Whilea high number of groups reduces the objective function it increases the potentialfor overfitting (Brusco et al., 2008). In order to find the optimal number of groupswe estimated our main regression for G = 1 until G = 10. The objective functiondecreases steadily as G increases: by around 50% when G = 2 compared toOLS, and by more than 80% when G = 8. Figure 512 reports the coefficients ofsocial and economic globalisation for different values of G. While the coefficientsfluctuate with few groups they become stable from G = 8. Given this sizeablereduction of the objective function and the stability of the coefficients for highernumbers of groups, we chose G = 8 as our optimal specification. Interestingly,the second to last row of tables 7,8, and 9 show that the objective function ofgrouped fixed effects is lower than the one of country and year fixed effects assoon as G ≥ 6 (G ≥ 7 for sugar). This suggests that a substantial amount ofcross-country heterogeneity is time-varying in our sample. The last row shows theobjective function when grouping countries into high, upper middle and, lowermiddle income countries. The value of the objective function is substantially largersuggesting that grouping according to income does not capture much of unobservedtime-varying heterogeneity.We conducted the same analysis for our health outcome variables and chose G= 10.Figure 8 in the Appendix plots the coefficient for different number of groups.

4 Results

This section reports estimation results for nutrition and health outcomes.

4.1 Nutrition outcomes

We find that social globalisation has a positive and significant effect on the shareof animal protein but no effect on free fat, and sugar in total food supply. Wefind no statistically significant effect of economic globalisation. These resultsare partly in line with the model predictions of Olivier et al. (2008). Our resultsconfirm the model prediction of a positive relationship between social globalisation

12 See tables 7,8, and 9 for the corresponding values.

14

4.1 Nutrition outcomes 4 RESULTS

Figure 5: Coefficients of economic and social globalisation

and cultural convergence but our empirical analysis does not reveal a significantnegative effect of economic globalisation on convergence.

Table 2 presents our main results for nutrition outcomes. Columns 1, 4, and 7 reportresults from a pooled OLS regression to facilitate comparison with Costa-i-Fontand Mas (2014). Our analysis confirms findings of Costa-i-Font and Mas (2014).Social globalisation is positively associated with the share of animal protein, freefat, and sugar in total food supply and the coefficient is significant at the 5% or 1%level. In contrast, the coefficient of economic globalisation is never significant.Results of country and year fixed effects regressions are presented in columns2,5, and 7. The coefficient of social globalisation remains positive but loses itssignificance for the share of free fat and sugar. The coefficient of economicglobalisation remains insignificant.Controlling for grouped time effects (columns 3, 6, and 9) the coefficient of socialglobalisation is positive for all outcome variables but only significant at the 1%level for animal protein. Compared to OLS and FE specifications it slightly shrinksin magnitude but remains to be of economic significance. An additional point onthe social globalisation index is associated with an increase in the share of animalprotein by 0.028. A one standard deviation increase (20.65) on the index of socialglobalisation increases on average the share of animal protein by about 0.6, whichis equivalent to an increase of about 12%.

15

4.2 Health outcomes 4 RESULTS

The effect of economic globalisation is negative for animal protein and free fat butthe coefficient is never significant for any of the outcome variables. The positivesign of coefficient of GDP per capita and the negative sign of the coefficient of itssquared term suggests a inverted U-shape of the relationship between GDP and theoutcome variables. The share of animal protein and free fat in the diet increaseswith increasing GDP up to a turning point beyond which increasing GDP reducesthe share of these components in the diet.

Table 2: Globalisation and nutrition outcomes

Outcome variable Animal protein Free fat SugarOLS FE GFE OLS FE GFE OLS FE GFE(1) (2) (3) (4) (5) (6) (7) (8) (9)

Social 0.031** 0.030***0.028*** 0.081** 0.021 0.022 0.211*** 0.017 0.039globalisation (0.012) (0.007) [0.010] (0.032) (0.021) [0.032] (0.054) (0.027) [0.035]Economic 0.011 0.001 -0.008 -0.036 -0.028 -0.045 -0.041 -0.024 0.020globalisation (0.009) (0.006) [0.012] (0.030) (0.022) [0.030] (0.051) (0.033) [0.036]GDP per capitaa 0.193*** 0.083** 0.174*** 0.471*** 0.102 0.454*** 0.085 0.240 0.168

(0.027) (0.041) [0.058] (0.115) (0.173) [0.131] (0.119) (0.164) [0.317](GDP per capita)2 -0.002*** -0.001** -0.002** -0.007*** -0.003 -0.007*** -0.003* -0.004* -0.003

(0.000) (0.001) [0.001] (0.002) (0.002) [0.002] (0.002) (0.002) [0.006]Constant 47.072***1.605***1.207*** -41.208 10.959*** 8.981*** 301.866***16.906***7.481***

(11.837) (0.215) (0.184) (50.957) (0.749) (0.687) (52.040) (1.261) (1.112)Country FE no yes no no yes no no yes noGroup FE no no yes no no yes no no yesYear FE no yes yes no yes yes no yes yesGroup-year FE no no yes no no yes no no yesTime trend yes no no yes no no yes no noN 2,940 2,940 2,940 2,940 2,940 2,940 2,940 2,940 2,940Adjusted R2 0.700 0.950 0.959 0.524 0.879 0.899 0.301 0.893 0.901Objective 4939 787.9 602.6 38657 9433 7274 77447 11439 9702

Robust standard errors in round brackets. Bootstrapped standard errors in square brackets. Results obtained withalgorithm 2.a GDP per capita (constant 2005 in 1000 USD).

4.2 Health outcomes

Results for nutrition outcomes are presented in table 3. We do not find significanteffects of social or economic globalisation on diabetes prevalence and mean BMI.Pooled OLS results suggest a positive and significant association between socialglobalisation and diabetes prevalence and mean BMI. But applying country andyear fixed effects (column 2) or grouped fixed effects (column 3) renders the coeffi-cients of social globalisation insignificant.Our results do not support findings of Miljkovic et al. (2015); Vogli et al. (2014);

16

4.3 Convergence across country groups 4 RESULTS

Costa-i-Font and Mas (2014) and Goryakin et al. (2015) who report a positiveand significant association between social globalisation and different measures ofoverweight.The coefficient of economic globalisation is only significant in the fixed effectsspecification. It suggests a negative effect of economic globalisation on diabetesprevalence and mean BMI. The coefficient however loses significance when apply-ing the GFE estimator.The values of the objective function of the GFE estimation is lower than the valuesof the objective function of the OLS and fixed effects estimation. This suggeststhat some cross-country heterogeneity is time-varying in our sample.

Table 3: Globalisation and health outcomes

Outcome variable Diabetes prevalence Mean BMIOLS FE GFE OLS FE GFE(1) (2) (3) (4) (5) (6)

Social globalisation 0.039** 0.003 -0.004 0.063*** -0.005 0.006(0.018) (0.015) [0.019] (0.014) (0.005) [0.010]

Economic globalisation -0.008 -0.038** -0.005 0.008 -0.015*** 0.013(0.012) (0.015) [0.013] (0.012) (0.005) [0.010]

GDP per capitaa -0.159*** -0.080 -0.063 -0.029 0.001 -0.005(0.038) (0.067) [0.082] (0.037) (0.020) [0.056]

(GDP per capita)2 0.001** 0.001 0.000 -0.000 -0.000 -0.000(0.001) (0.001) [0.001] (0.001) (0.000) [0.001]

Constant -121.998*** 9.498*** 7.963*** -32.201 24.738*** 19.780***(24.616) (0.272) (0.224) (25.502) (0.154) (0.277)

Country FE no yes no no yes noGroup FE no no yes no no yesYear FE no yes yes no yes yesGroup-year FE no no yes no no yesTime trend yes no no yes no noN 2,494 2,494 2,494 2,494 2,494 2,494Adjusted R2 0.328 0.874 0.925 0.439 0.973 0.972Objective 6681 1191 656.1 4398 198.6 193.3

Robust standard errors in round brackets. Bootstrapped standard errors in square brackets.Results obtained with algorithm 2.a GDP per capita (constant 2005 in 1000 USD).

4.3 Convergence across country groups

Moreover, we are addressing the question of whether countries are convergingtowards a homogeneous nutritional profile. Figures 1 and 2 already suggested thatmiddle income countries have not yet reached the high levels of animal proteinand free fat in the diet, which are prevalent in high income countries. But is the

17


gap shrinking between middle and high income countries? Figure 6 shows thecoefficient of variation13 over time for the nutrition outcome variables. The shareof sugar in the diet exhibits a strong convergence trend over time. The coefficientof variation moderately declines for animal protein, while it is increasing for freefat. This suggests that middle income countries are catching up to high incomecountries regarding the share of sugar in their diet, and to a smaller extent withrespect to animal protein.

Figure 6: Coefficient of variation for nutrition outcomes, 1970-2011

The observed convergence for sugar and animal protein can be explained to a largedegree by globalisation and GDP. Figure 7 plots the difference between groupsover time. The difference is calculated as the sum of the group dummy and thegrouped time effect14. We find that after controlling for globalisation and GDP,the difference across country groups appears stable for animal protein indicatingthat the moderate convergence process of animal protein can be explained byglobalisation and GDP. But for sugar, we observe a small convergence trendafter controlling for globalisation and GDP. This suggests that other factors partlyexplain the observed strong convergence of countries with respect to the share ofsugar in their diet.

Finally, we investigate whether social globalisation has a different effect on thediet for different groups. To this end, we interacted the variable social globalisationwith the group indicator variables.Table 4 present the impact of social globalisation for different groups. For animal

13 the coefficient of variation is defined as the ratio of the standard deviation to the mean.14 e.g. for group 3 and year 1971 it is the sum of group3 and group3 ∗ year1971 relative to the basegroup, group 1.

18


Figure 7: Convergence of groups for nutrition outcomes, 1970-2011

protein (column 1), the coefficient of the interaction term is positive and significantfor four groups (groups 1,2,7, and 8). Group 2 exhibits the largest coefficient andthis group consists of 14 mostly high income and upper middle income countries(Chile, Denmark, Portugal, China, Colombia, Costa Rica, Dominican Republic,Fiji, Malaysia, Mexico and the lower middle income countries Honduras, Kenya,Pakistan, and Senegal). In contrast, the coefficient of group 1 is much smaller andthis group consists of 15 mostly lower middle income countries (Algeria, Tunisia,Turkey, Cameroon, Cote d‘Ivoire, Egypt, El Salvador, Guatemala, India, Indonesia,Lesotho, Morocco, Nigeria, Zambia, and the high income country Saudi Arabia).Groups 7 and 8 each consists only of one country, namely Greece and Israel. Thesefindings suggest that the effect of social globalisation on the share of animal proteinin the diet is stronger for richer than for poorer countries. This corresponds to ourhypothesis of a a moderate convergence process between middle and high incomecountries with respect to animal protein in the diet. For the outcome variables freefat and sugar we do not find any significant effects of the interaction terms.

One potential concern is that these results do not capture a different impact of socialglobalisation for different groups but simply reflect that different groups exhibitsubstantial differences in their median level of social globalisation. In other wordsthe groups may just be at different stages of social globalisation. The box plotin figure 9 mitigates this concern, as group 1 and 2 do not exhibit a substantiallydifferent level of social globalisation compared to the other groups.

19

5 DISCUSSION

Table 4: Heterogeneous effects by group, grouped fixed effects

Outcome variable Animal proteina Free fata Sugara

(1) (2) (3)Social globalisation*group 8 0.074*** 0.072 -0.022

(0.024) (0.095) (0.114)Social globalisation*group 7 0.049** 0.096 0.347

(0.023) (0.105) (0.252)Social globalisation*group 6 0.031 0.057 -0.081

(0.036) (0.074) (0.251)Social globalisation*group 5 0.038 -0.006 0.174

(0.024) (0.094) (0.371)Social globalisation*group 4 0.042 0.006 0.101

(0.142) (0.12) (0.104)Social globalisation*group 3 0.006 -0.213 0.083

(0.03) (0.153) (0.104)Social globalisation*group 2 0.130*** 0.022 0.011

(0.05) (0.123) (0.119)Social globalisation*group 1 0.023* -0.131 -0.237

(0.013) (0.204) (0.348)Economic globalization 0.007 0.015 0.027

(0.016) (0.07) (0.118)GDP per capitab 0.179*** 0.413* 0.144

(0.052) (0.242) (0.184)(GDP per capita)2 -0.002*** -0.005 -0.002

(0.001) (0.004) (0.003)N 2940 2940 2940

Results obtained with algorithm 1 and 500 randomly generated starting values. Clusteredstandard errors based on the large-T normal approximation in parentheses. Regressionsinclude group FE, year FE and, group-year FE.a The dependent variable is the share of animal protein (free fat, sugar) in total food supply.b GDP per capita (constant 2005 in 1000 USD).

5 Discussion

Our results suggest that social globalisation has a positive effect on the shareof animal protein in the diet while economic globalisation has no effect on thenutrition composition. This result is only partly in line with the model developedby Olivier et al. (2008) predicting a positive effect of social globalisation but anegative effect of economic globalisation on cultural convergence.The next step of our analysis consists of analysing the impact of the different sub-components of the globalisation index in order to better understand the underlyingmechanisms.

20

5 DISCUSSION

Table 5 presents results from estimating the effect of different subcomponentsof the globalisation index on the share of animal protein, free fat, and sugar intotal food supply. We find that the positive effect of social globalisation seemsto be mostly driven by information flows (e.g. number of Internet and telephoneusers per 1000 people), as this sub-component has the largest coefficient and isalways positive and significant at the 1% level. Interestingly, personal contacts (e.g.tourism, telephone traffic, foreign population and transfers) have a positive effecton the share of animal protein in the diet but a negative effect on free fat and sugar.Countries with a high score on personal contacts are mostly small countries witha relatively large share of foreign populations (e.g. Ireland, Austria, Netherlands,Denmark, Norway). The foreigners may quickly adjust their meat consumption butonly slowly alter their dietary habits with respect to free fat and sugar explainingthe negative effect of personal contacts on the share of free fat and sugar in the diet.

Finally, cultural proximity (e.g. number of McDonald’s restaurants and IKEA)has a small positive effect on free fat and a larger negative effect on the shareof sugar in the diet. Countries scoring high on the cultural proximity index arehigh income countries such as Australia, United States, Austria, Denmark and theUnited Kingdom. One potential explanation is that we do not adequately accountfor sugar production made from corn in our food item classification. 80% of thisso-called high-fructose corn syrup is produced in rich countries such as the US,Japan, the EU, and South Korea and is widely used as a sweetener in the food andbeverage industry (Credit Suisse Research Institute, 2013).Overall, the effects of the subcomponents for free fat and sugar carry positive aswell as negative signs and thus are likely to cancel each other out in the compositesocial globalisation variable. This could explain why we do not find significanteffects of social globalisation on the share of free fat and sugar in the diet.

21

6 CONCLUSION

Table 5: Subcomponents of social globalisation, grouped fixed effects

Outcome variable Animal protein Free fat Sugar(1) (2) (3)

Personal contacts 0.008*** -0.027*** -0.040***(0.002) (0.009) (0.011)

Information flows 0.026*** 0.035*** 0.124***(0.002) (0.009) (0.012)

Cultural proximity 0.002 0.009* -0.018***(0.001) (0.005) (0.006)

Economic globalisation -0.000 -0.020 0.017(0.002) (0.012) (0.011)

GDP per capitaa 0.172*** 0.457*** -0.021(0.008) (0.029) (0.035)

(GDP per capita)2 -0.002*** -0.006*** 0.000(0.000) (0.001) (0.001)

Constant 0.557*** 10.199*** 6.242***(0.201) (0.587) (1.272)

N 2,940 2,940 2,940Adjusted R2 0.962 0.902 0.914

Bootstrapped standard errors in parentheses. Results obtained with algorithm 2.Regressions include group FE, year FE and, group-year FE.a GDP per capita (constant 2005 in 1000 USD).

6 Conclusion

This paper contributes to explaining the causes of changing nutrition patterns byestimating the effect of globalisation on the composition of the diet and healthoutcomes. Globalisation has often be held responsible for changing diets andassociated negative health outcomes such as obesity, diabetes, and cardiovasculardiseases. However, existing evidence concentrated on economic aspects of globali-sation and mostly consisted of case studies about trade liberalisations and FDI.We provide empirical evidence on the impact of globalisation on the share ofanimal protein, free fat, and sugar in food supply as well as on diabetes and meanBMI by using a panel of 70 high and middle income countries from 1970 until2011. In order to better account for unobserved time-varying heterogeneity weapply a grouped fixed effects estimator developed by Bonhomme and Manresa(2015). This estimator endogenously groups countries together by minimizing aleast-squares criterion and subsequently controls for group-specific time-varyingunobserved heterogeneity.Our results indicate that the social dimension of globalisation has a positive andsignificant effect on the share of animal protein in the diet while economic glob-

22

6 CONCLUSION

alisation has no impact on the composition of the diet. A one standard deviationincrease on the index of social globalisation increases the share of animal proteinon average by 12%. This finding is relevant for economies given the negativehealth consequences of a meat-intensive diet and associated healthcare costs aswell as the environmental impact of meat production.Moreover, we find that the gap between country groups has strongly convergedover time for sugar and to a lesser extent for animal protein. For animal protein theobserved convergence trend can be attributed to globalisation and GDP while inthe case of sugar additional factors must play a role that require further analysis.We further show that the effect of social globalisation on animal protein is strongerfor richer than for poorer countries, supporting the hypothesis of a moderate con-vergence trend between high and middle income countries with respect to the shareof meat in their diet.Given that social globalisation seems to be the main driver of changing diets, furtherresearch should focus on factors related to the social dimension of globalisation,such as food advertising on television and the Internet.

23

REFERENCES REFERENCES

References

Al-Rodhan, N. R. and G. Stoudmann (2006). Definitions of globalization: Acomprehensive overview and a proposed definition. Program on the GeopoliticalImplications of Globalization and Transnational Security, 6.

Bishwajit, G., S. Ide, M. A. Hossain, and M. N. Safa (2014). Trade liberalization,urbanization and nutrition transition in asian countries. Journal of NutritionalHealth & Food Science, 2(1).

Bonhomme, S. and E. Manresa (2015). Grouped patterns of heterogeneity in paneldata. Econometrica, 83(3).

Brusco, M. J., J. D. Cradit, D. Steinley, and G. L. Fox (2008). Cautionary remarkson the use of clusterwise regression. Multivariate Behavioral Research, 43(1):29–49.

Costa-i-Font, J. and N. Mas (2014). ’Globesity’? The Effects of Globalization onObesity and Caloric Intake. CESifo Working Paper Series, 4982.

Credit Suisse Research Institute (2013). Sugar - consumption at crossroads. Tech-nical report, Credit Suisse.

de Souza, R. J., A. Mente, A. Maroleanu, A. I. Cozma, V. Ha, T. Kishibe, E. Uleryk,P. Budylowski, H. Schünemann, J. Beyene, and S. S. Anand (2015). Intakeof saturated and trans unsaturated fatty acids and risk of all cause mortality,cardiovascular disease, and type 2 diabetes: systematic review and meta-analysisof observational studies. BMJ, 351.

Dreher, A. (2006). Does globalization affect growth? evidence from a new indexof globalization. Applied Economics, 38(10):1091–1110.

Finucane, M. M., G. A. Stevens, M. J. Cowan, G. Danaei, J. K. Lin, C. J. Paciorek,G. M. Singh, H. R. Gutierrez, Y. Lu, A. N. Bahalim, F. Farzadfar, L. M. Riley, andM. Ezzati (2011). National, regional, and global trends in body-mass index since1980: systematic analysis of health examination surveys and epidemiologicalstudies with 960 country-years and 9Â·1 million participants. The Lancet,377(9765):557 – 567.

Folkvord, F., D. J. Anschütz, E. Boyland, B. Kelly, and M. Buijzen (2016). Foodadvertising and eating behavior in children. Current Opinion in BehavioralSciences, 9:26 – 31. Diet, behavior and brain function.

Goryakin, Y., T. Lobstein, W. P. T. James, and M. Suhrcke (2015). The impact ofeconomic, political and social globalization on overweight and obesity in the 56low and middle income countries. Social Science & Medicine, 133:67 – 76.

Hawkes, C. (2006). Uneven dietary development: linking the policies and processesof globalization with the nutrition transition, obesity and diet-related chronicdiseases. Globalization and Health, 2(1):1–18.

24


Hawkes, C. and A. M. Thow (2008). Implications of the central america-dominicanrepublic-free trade agreemeent for the nutrition in central america. Pan AmericanJournal of Public Health, 24(5):345–360.

Imamura, F., L. O’Connor, Z. Ye, J. Mursu, Y. Hayashino, S. N. Bhupathiraju, andN. G. Forouhi (2015). Consumption of sugar sweetened beverages, artificiallysweetened beverages, and fruit juice and incidence of type 2 diabetes: systematicreview, meta-analysis, and estimation of population attributable fraction. BMJ,351.

Leren, P. (1968). The effect of plasma-cholesterol-lowering diet in male survivorsof myocardial infarction. a controlled clinical trial. Bulletin of the New YorkAcademy of Medicine, 44(8):1012.

Malik, V. S., Y. Li, D. K. Tobias, A. Pan, and F. B. Hu (2016). Dietary proteinintake and risk of type 2 diabetes in us men and women. American Journal ofEpidemiology, 183(8):715–728.

Miljkovic, D., S. Shaik, S. Miranda, N. Barabanov, and A. Liogier (2015). Globali-sation and obesity. The World Economy, 38(8):1278–1294.

Monteiro, C. A., J.-C. Moubarac, G. Cannon, S. W. Ng, and B. Popkin (2013).Ultra-processed products are becoming dominant in the global food system.Obesity Reviews, 14:21–28.

Olivier, J., M. Thoenig, and T. Verdier (2008). Globalization and the dynamics ofcultural identity. Journal of International Economics, 76(2):356–370.

Pettigrew, S., L. Tarabashkina, M. Roberts, P. Quester, K. Chapman, and C. Miller(2013). The effects of television and internet food advertising on parents andchildren. Public health nutrition, 16(12):2205–2212.

Popkin, B. M. (2006). Technology, transport, globalization and the nutritiontransition food policy. Food Policy, 31(6):554 – 569.

Reardon, T., C. P. Timmer, C. B. Barrett, and J. Berdegué (2003). The rise ofsupermarkets in africa, asia, and latin america. American Journal of AgriculturalEconomics, 85(5):1140–1146.

Stuckler, D., M. McKee, S. Ebrahim, and S. Basu (2012). Manufacturing epidemics:The role of global producers in increased consumption of unhealthy commoditiesincluding processed foods, alcohol, and tobacco. PLoS Med, 9(6):1–8.

Te Morenga, L., S. Mallard, and J. Mann (2013). Dietary sugars and body weight:systematic review and meta-analyses of randomised controlled trials and cohortstudies. BMJ, 346.

Thow, A. M. and C. Hawkes (2009). The implications of trade liberalization fordiet and health: a case study from central america. Globalization and Health,5(1):1–15.

Thow, A. M., P. Heywood, J. Schultz, C. Quested, S. Jan, and S. Colagiuri (2011).Trade and the nutrition transition: Strengthening policy for health in the pacific.Ecology of Food and Nutrition, 50(1):18–42.

25


Vogli, R. D., A. Kouvonen, M. Elovainio, and M. Marmot (2014). Economicglobalization, inequality and body mass index: a cross-national analysis of 127countries. Critical Public Health, 24(1):7–21.

WHO (2000). Obesity: preventing and managing the global epidemic, number894. World Health Organization.

WHO (2015a). Cardiovascular diseases. Fact Sheet, 317.WHO (2015b). Obesity. Fact Sheet, 311.

26

7 APPENDIX

7 Appendix

Table 6: Summary statistics, health sample

Mean SD Min MaxOutcome variables

Diabetes prevalence 8.39 2.00 3.18 21.82Mean BMI 24.57 1.78 19.47 28.74

CovariatesEconomic globalisation 54.69 18.11 9.94 99.16Social globalisation 46.92 21.28 6.85 93.68GDP per capitaa 11.75 14.80 0.22 86.13

Income groupsHigh income 0.42 0.49 0.00 1.00Upper middle income 0.29 0.45 0.00 1.00Lower middle income 0.29 0.45 0.00 1.00

SampleNumber of countries 86Number of years 29N 2494

a GDP per capita (constant 2005 in 1000 USD).

Figure 8: Coefficients of economic and social globalisation

27

7 APPENDIX

Table 7: GFE estimates, animal protein

Number Objective Social Economic GDP (GDPof groups globalisation globalisation per capitaa per capita)2

1 4918 0.033** 0.011 0.189*** -0.002***(0.013) (0.009) (0.029) (0.000)

2 2506 0.032** -0.021 0.193*** -0.002**(0.015) (0.016) (0.044) (0.001)

3 1350 0.035*** 0.007 0.204*** -0.003***(0.006) (0.007) (0.033) (0.001)

4 1102 0.037*** 0.008 0.214*** -0.003***(0.006) (0.011) (0.056) (0.001)

5 868.2 0.024*** -0.006 0.157*** -0.002**(0.008) (0.010) (0.051) (0.001)

6 764.9 0.020*** -0.003 0.156*** -0.002**(0.011) (0.010) (0.026) (0.000)

7 699.7 0.030** -0.011 0.164*** -0.002**(0.012) (0.011) (0.046) (0.001)

8 602.6 0.028*** -0.008 0.174*** -0.002**(0.010) (0.012) (0.058) (0.001)

9 547.4 0.028*** -0.009 0.172*** -0.002**(0.010) (0.008) (0.045) (0.001)

10 516 0.029*** -0.01 0.171*** -0.002**(0.009) (0.009) (0.041) (0.001)

Country & year FE 787.9 0.030*** 0.001 0.083** -0.001**(0.007) (0.006) (0.041) (0.001)

Income groupsb 4223 0.033** 0.003 0.080 -0.001(0.012) (0.010) (0.055) (0.001)

The table reports the value of the objective function and the GFE coefficient for variousnumber of groups. Computation using algorithm 2 (5;10;5) with 25 Bootstrap iterations.a GDP per capita (constant 2005 in 1000 USD).b Countries grouped into high income, upper middle, and lower middle income countries.

28

7 APPENDIX

Table 8: GFE estimates, free fat


1 38110 0.090*** -0.037 0.447*** -0.007***(0.033) (0.030) (0.118) (0.002)

2 17781 0.049 -0.052 0.431*** -0.007***(0.031) (0.035) (0.060) (0.001)

3 14011 0.035 -0.022 0.399*** -0.006***(0.031) (0.042) (0.093) (0.002)

4 11546 0.047 -0.092** 0.410*** -0.006**(0.032) (0.043) (0.147) (0.002)

5 9739 0.026 -0.066* 0.432*** -0.006***(0.021) (0.036) (0.121) (0.002)

6 8603 0.026 -0.063 0.427*** -0.006***(0.032) (0.046) (0.111) (0.002)

7 7861 0.015 -0.013 0.450*** -0.006***(0.027) (0.038) (0.122) (0.002)

8 7274 0.022 -0.045 0.454*** -0.007***(0.032) (0.030) (0.131) (0.002)

9 6726 0.032 -0.052* 0.333*** -0.003(0.030) (0.027) (0.113) (0.002)

10 6174 0.029 -0.047 0.333* -0.003(0.025) (0.034) (0.144) (0.003)

Country & year FE 9433 0.021 -0.028 0.102 -0.003(0.021) (0.022) (0.173) (0.002)

Income groupsb 35583 0.092*** -0.048 0.351** -0.005**(0.034) (0.031) (0.137) (0.002)


29

7 APPENDIX

Table 9: GFE estimates, sugar


1 76684 0.223*** -0.041 0.055 -0.003(0.056) (0.052) (0.124) (0.002)

2 36924 0.138** -0.034 0.373** -0.006***(0.068) (0.092) (0.169) (0.003)

3 24074 0.108** 0.038 0.378* -0.006***(0.056) (0.048) (0.218) (0.003)

4 17458 0.079* 0.04 0.607** -0.009***(0.043) (0.040) (0.257) (0.004)

5 13690 0.056 0.022 0.085 -0.001(0.038) (0.039) (0.213) (0.003)

6 12085 0.063* 0.014 0.335** -0.005***(0.037) (0.031) (0.164) (0.003)

7 10727 0.06 0.015 0.319 -0.005(0.051) (0.028) (0.282) (0.005)

8 9702 0.039 0.02 0.168 -0.003(0.035) (0.036) (0.317) (0.006)

9 8747 0.028 0.016 0.134 -0.003(0.043) (0.043) (0.202) (0.003)

10 7909 0.041 0.022 0.029 -0.000(0.046) (0.038) (0.312) (0.005)

Country & year FE 11439 0.017 -0.024 0.240 -0.004*(0.027) (0.033) (0.164) (0.002)

Income groupb 69468 0.230*** -0.066 -0.117 0.000(0.050) (0.052) (0.133) (0.002)


Figure 9: Distribution of social globalisation across groups

30

Globalisation and national trends in nutrition and health ... · Globalisation has substantially...

Documents

Transcript of Globalisation and national trends in nutrition and health ... · Globalisation has substantially...